World Income Inequality Database (WIID) Version 4. User Guide and Data Sources

Transcription

1 World Income Inequality Database (WIID) Version 4 User Guide and Data Sources 19 December 2018

2 Contents Preface... iii Basic principles behind WIID Conceptual base... 1 Income or consumption?... 1 Income concept... 1 Consumption/expenditure concept... 2 Other conceptual issues... 2 Information regarding OECD, Eurostat, LIS, World Bank and SEDLAC databases... 4 Construction of WIID New observations... 6 Corrected observations... 6 Approach to the grouped variables... 7 Variable level changes... 7 Database format... 8 Documentation... 9 Documentation in the database itself... 9 Country information sheets Quality rating Criteria used Final rating Quality score Some final guidelines Referring to the WIID List of variables Glossary Appendix A: Codebook Appendix B: United Nations geographical sub-regions Appendix C: World Bank regional classification Sources References ii

3 Preface The UNU-WIDER World Income Inequality Database widely known by its acronym WIID provides information on income inequality for 189 developed, developing, and transition countries (including historical entities) in an organized and user-friendly manner. Two decades ago the first version of the database was compiled. Since then it has been subsequently updated and it has expanded significantly. The current version WIID4 is the fourth major edition of the database. It is part of the UNU-WIDER work programme on Transformation, Inclusion, and Sustainability. WIID4 retains several elements of the concept of the previous version, WIID3.4, but it also includes many changes, both in the content and structure of the variables. The database contains more than 11,000 observations, now reaching the year 2017, covering almost every country in the world. Finally, the new version also corrects for inconsistencies and other issues found in earlier editions. The latest update was prepared by a UNU-WIDER team including Antti Pelanteri, Carlos Gradín Lago, Anustup Kundu and Heini Salonen, and greatly benefited from the contributions by Tony Shorrocks and Çınar Baymul. With this new version, WIID continues to be the most comprehensive source of data on income inequality in the world. We trust that it will continue to serve the global research community and members of the public interested in describing inequality in specific countries and analyzing global inequality trends. In addition, the database has also been extensively used for research on a broad range of socio-economic issues related to economic inequality. We warmly welcome all user feedback and suggestions to further enhance the quality of the WIID. Finn Tarp Director, UNU-WIDER Helsinki, Finland December 2018 iii

4 Basic principles behind WIID4 Conceptual base Unlike national accounts data which are in principle comparable across countries, there is no agreed basis of definition for the construction of distribution data. Sources and methods might vary, especially across but also within countries. This may be the case even if the data comes from the same source. In their influential articles on the use of secondary data in studies of income distribution, Atkinson and Brandolini (2001, 2009) discuss quality and consistency in income distribution data both within and across countries. They show how both levels and trends in distributional data can be affected by data choices. In light of this, it is not an easy task to construct a secondary database with distribution data. To get some structure, we started by defining a preferred set of features for the conceptual base and the underlying data. With the conceptual base we mean the definitions of income or consumption/expenditure, the statistical units to be adopted, the use of equivalence scales and weighting. Income or consumption? The first issue to address is whether inequality estimates based on income or consumption should be preferred. According to Deaton and Zaidi (2002), the empirical literature on the relationship between income and consumption has established, for both rich and poor countries, that consumption is not closely tied to short-term fluctuations in income, and that consumption is smoother and less variable than income. Especially in developing countries, where the rural agriculture sector is large, it is difficult to gather accurate income data. Accordingly, consumption data should be used. Atkinson and Bourguignon (2000) do not share this view. There is, according to them, no clear advantage in using consumption rather than income in studying distributional issues. The use of consumption rather than income data raises problems of definition and observation, the main conceptual problem being the treatment of durables and the necessity of imputing value for their services. Regardless of the different views, the collection of inequality observations is restricted to what in practice is available. In most industrialized countries, inequality and poverty are assessed with reference to income not consumption (Deaton and Zaid 2002). This tradition is followed in much of Latin America. By contrast, most Asian and African surveys have always collected detailed consumption data. The fact that distribution data can be based on both income and consumption is the first stepping stone in the construction of comparable statistics. In WIID4 we have strived to collect observations with reference to both income and consumption, whenever it is possible. Income concept The second issue is how to define income and consumption. As stated earlier, there is no agreed basis of definition as in the case of national accounts data. Concerning income data, some steps have been taken towards developing international standards. The Final Report and Recommendations of the Canberra Group (2001) provides an appropriate base for defining the most preferred income concept as the objective of the group was to enhance national household income statistics by developing standards on conceptual and practical issues related to the production of income distribution statistics. Even if the work of the group is mainly based on OECD-country experience, we believe that the main conclusions concerning the income concept also hold for other countries. In Table 1, the income concept as recommended by the Canberra Group for international comparisons of income distribution is given. The definition of total and disposable income as recommended by the group should include certain components to be considered complete. We have been drawing special attention to whether the 1

5 underlying income concept includes income items such as imputed rents for owner-occupied dwellings, 1 imputed incomes from home production and in-kind income in general. Imputed rent from owneroccupied dwellings is not mentioned in the concept of the Canberra Group since many countries do not provide estimates for this item, and it is differently valued in different countries. Imputed rents should, however, preferably be included even if the comparability between countries might suffer somewhat. Home production and in-kind income are crucial in developing and transition countries. The income concept cannot be considered complete for these countries if income in-kind and income from home production are not included. The inequality indices reported will in the first place be those calculated on the basis of disposable income, but if indices based on earnings or gross incomes (total income, according to the Canberra Group terminology) are available, they will also be reported. Consumption/expenditure concept On the consumption side, the situation is more difficult. Deaton and Zaidi (2002) from the LSMS group at the World Bank 2 have worked out some guidelines. Their recommendations on how to use consumption data for welfare measurement were used. Where the Canberra Group recommendations were built mainly on OECD-country experience, these recommendations are mainly built on experiences from developing countries. The crucial thing here is to evaluate the consumption rather than to simply calculate the expenditures. In other words, to make a distinction between what is consumed and what is purchased. This means that one is not interested in the purchase value of durable goods but in the use or rental value. As is clear from Table 1, taxes paid, purchase of assets, repayments of loans, and lumpy expenditures should not be included in the consumption aggregate. If they are included, we refer to expenditure rather than consumption. Again, we have paid attention to the inclusion of non-monetary items. Other conceptual issues The third issue to look at concerns other conceptual issues. Here we follow quite closely the recommendations of the Canberra Group. Departures from the recommendations are mainly driven by practical matters. a) The household should be the basic statistical unit; the statistical unit for analysis of economic well-being has to be one where assumptions of sharing of economic resources are most plausible. The Canberra Group motivates the preference for the household by the relationship of households to both micro (survey) and macro (SNA) data uses. In practice, households are often used as the basic statistical unit. The different definitions of households that appear in the data are a problem which will affect the estimates and users should be aware of. b) Income or consumption should be adjusted to take account of household size, using per capita incomes or consumption. The Canberra Group suggests the use of equivalence scales as the relative need of different sized households is different. We decided to choose per capita estimates as the preferred ones, as they are the one mostly commonly available and since a lot of different equivalence scales are in use which weakens the comparability of the estimates. 1 Please refer to the glossary for an explanation of the terms used. 2 LSMS stands for Living Standards Measurement Study. The household surveys provided by this study can be found at: K: ~piPK: ~theSitePK: ,00.html 2

6 Table 1: Preferred set of underlying concepts for inequality estimates in WIID4 The income concept recommended by the Canberra Group for international comparisons of income distribution: 1. Employee income Cash wages and salaries 2. Income from self-employment Profit/loss from unincorporated enterprise Imputed income from self-employment Goods and services produced for barter, less cost of inputs Goods produce for home consumption, less cost of inputs 3. Income less expenses from rentals, except rent of land 4. Property Income Interest received less interest paid Dividends 5. Current transfers received Social insurance benefits from employers schemes Social insurance benefits in cash from government schemes Universal social assistance benefits in cash from government Means-tested social assistance benefits in cash from government Regular inter-household cash transfers received 6. Total income (sum of 1 to 5) 7. Current transfers paid Employees social contributions Taxes on income 8. Disposable income (6 less 7) Other conceptual issues: The consumption aggregate recommended by Deaton & Zaidi (2002) for welfare measurements: 1. Food consumption Food purchased from market Home produced Received as gift or in-kind payment 2. Non-food consumption Daily use items Clothing and houseware Health expenses Education expenses Transport 3. Durable goods The use-value (rental value) of durables 4. Housing Rents paid If dwelling is owned by household or received free of charge, an estimate of the rental equivalent (imputed rent) Utilities (water, electricity, garbage collection etc.) To be excluded: Taxes paid, purchase of assets, repayments of loans and lumpy expenditures. If durables are included with their purchase value or/and taxes paid, purchase of assets, repayments of loans and lumpy expenditures, the concept to be referred to is expenditures. 1. Household should be the basic statistical unit 2. Per capita incomes or consumption/expenditure should be measured 3. Person weights should be applied

7 c) Person weights are preferred as the users of income statistics most often are concerned with the economic well-being of individuals and not with the well-being of households. Estimates not following the preferred set of definitions are not automatically considered to be of bad quality, but when updates were made, the definitions were followed whenever we could make a choice. Due to unavailability of observations using the preferred set of definitions, estimates based on other definitions were in several cases used. The differences appear especially in the statistical units and in the weighting. Information regarding OECD, Eurostat, LIS, World Bank and SEDLAC databases WIID combines information coming from many sources, including historical compilations with updated information from the most salient data repositories (including LIS, SEDLAC, Eurostat, World Bank, OECD and ECLAC), as well as from national statistical offices, and independent research papers. In below we introduce the main data sources. OECD The Organisation for Economic Co-operation and Development (OECD) Income Distribution database (IDD) 3 has been developed to benchmark and monitor countries performance in the field of income inequality and poverty. It contains a number of standardized indicators based on the central concept of equivalised household disposable income ; i.e. the total income received by households less the current taxes and transfers they pay, adjusted for household size with an equivalence scale. While household income is only one of the factors shaping people s economic wellbeing, it is also the one for which comparable data for all OECD countries are most common. Income distribution has a long-standing tradition among household-level statistics, with regular data collections going back to the 1980s (and sometimes earlier) in many OECD countries. Achieving comparability in this field is a challenge, as national practices differ widely in terms of concepts, measures, and statistical sources. In order to maximize international comparability as well as inter-temporal consistency of data, the IDD data collection and compilation process is based on a common set of statistical conventions (e.g. on income concepts and components). The information obtained by the OECD through a network of national data providers, via a standardized questionnaire, is based on national sources that are deemed to be most representative for each country. Eurostat The EU-Statistics on Income and Living Conditions (EU-SILC) instrument is the EU reference source for comparative statistics on income distribution and social inclusion at the European level. It provides two types of annual data for 28 European Union countries, Iceland, Norway, Switzerland, and Turkey: Cross-sectional data pertaining to a given time or a certain time period with variables on income, poverty, social exclusion and other living conditions, and 3 and 4

8 Longitudinal data pertaining to individual-level changes over time, observed periodically over a four-year period. EU-SILC does not rely on a common questionnaire or a survey but on the idea of a framework. The latter defines the harmonized lists of target primary (annual) and secondary (every four years or less frequently) variables to be transmitted to Eurostat; common guidelines and procedures; common concepts (household and income) and classifications aimed at maximizing comparability of the information produced. The minimum size of the sample of the overall population which is surveyed every year is of: Cross-sectional data operation: about 130,000 households and 270,000 persons aged 16 and over are interviewed in the European Union countries. Longitudinal data operation: about 100,000 households and 200,000 persons aged 16 and over are interviewed in the European Union countries. The reference population in EU-SILC includes all private households and their current members residing in the territory of the countries at the time of data collection. Persons living in collective households and in institutions are generally excluded from the target population. Some small parts of the national territory amounting to no more than 2% of the national population and the national territories listed below may be excluded from EU-SILC. All household members are surveyed, but only those aged 16 and more are interviewed. LIS The Luxembourg Income Study (LIS) is the largest available income database of harmonized microdata and is based at the LIS Cross-National Data Center in Luxembourg. It mostly refers to developed economies, but it is increasingly expanding to incorporate middle income countries and, in the near future, low-income countries. This database is widely recognized as the main international reference for cross-country comparisons for the countries and years covered. The observations from LIS in WIID are directly provided by the LIS Cross-National Data Center. World Bank The World Bank provides an online tool, PovcalNet, to allow for country-level data estimation on-demand. The underlying concepts of the data acquired are difficult to track and hence we have graded the data mostly as average in our quality rating, but nevertheless this is an important data source given its impressive coverage across countries. SEDLAC The Socio-Economic Database for Latin America and the Caribbean (SEDLAC), based in CEDLAS (La Plata, Argentina) in collaboration with the World Bank, is a harmonized set of indicators based on a collection of surveys. It has increasingly been considered as the main reference for cross-country inequality comparisons in the region. The WIID has acquired data directly from CEDLAS. 5

9 Construction of WIID4 The data points in a secondary database will originate from different sources and refer to a variety of income and population concepts, sample sizes, and statistical methods. To deal with this reality the only thing one can do, is to specify as precisely as possible the conceptual base for each observation and to also otherwise document the data well. Atkinson and Brandolini (2001), Pyatt (2003), and Székeley and Hilgert (1999), who are critical of the use of secondary databases, point in particular to the problem of insufficient documentation. This criticism was taken into account in the construction of WIID2 (see the User guide for WIID2, available from UNU-WIDER website). Jenkins (2015) provided a thorough review of WIID3 with suggestions on how it should be developed; Badgaiyan et al. (2015) addresses Jenkin s comments in detail. WIID4 retains several elements of the concept of the previous version, WIID3.4, but it also includes many changes, both in the content and structure of the variables. We have tried to report as thoroughly as possible the underlying data. New observations WIID4 comprises of 11,101 observations, whereas WIID3.4 had 8,817. The following summarizes the number of observations for different time periods: Time span Number of observations Total observations 11,101 Before , , , ,262 WIID now contains distributional data for 189 countries (including historical entities), up from the 182 in WIID3.4. The new country-year observations come from a number of sources: (1) household survey statistics obtained from national statistical offices of the corresponding countries; (2) the Socio-Economic Database for Latin America and the Caribbean (SEDLAC); (3) LIS Cross- National Data Center; (4) the Organisation for Economic Co-operation and Development (OECD); (5) Eurostat; (6) the World Bank s PovcalNet and from research outputs such as journal articles (7). Corrected observations WIID has been assembled from different sources, many dating back to times when paper records were the norm and transcription errors sometimes occurred. As a consequence, it included some duplicate observations, which have been eliminated, or coding errors and 6

10 mistakes that have been corrected. Some cases where the reported values of the Gini coefficient were inconsistent with historical trends have been verified with the source and corrected accordingly. Approach to the grouped variables The approach to the consolidated variables and the original full variables has been changed. Previously, information was provided fully in the original variables (e.g. AreaCovr) and then in a consolidated manner in the variables with the suffix _new (e.g. AreaCovr_new). Now, the logic has been altered to work to the other direction instead. In WIID4 the main variable, also by name, is the grouped variable, e.g. scale, and then the full is given in the detailed variable, e.g. scale_detailed. This change has been applied as it is convenient for most users to just use the consolidated variables. Categories for the grouped variables have been considerably reduced. This will allow most users to work immediately with the categories provided without creating their own mapping. Categories for the detailed variables have been reduced and edited. In some cases this has been done by checking from the source and in some cases removing the duplicate or near-duplicate categories. For the grouped and detailed variables, it is now much easier to follow which detailed values fall under which consolidated variable categories as the numerical values in the detailed variables are now clearly referring (with the first digit) to under which grouped category they fall into. Variable level changes All variable names have been made lowercase and variable labels have been updated. The order of the variables has also been changed. An identifier variable id has been added to the data. Country code variables are now named c3 and c2 respectively. The Gini coefficient variable is now named gini_reported, to highlight the fact that it is presented as in the originating source. Variables for the bottom five and top five percent of income earners have now been named bottom5 and top5 respectively, to avoid any confusion that the old convention (P5 and P95) might introduce. Previous variables for welfare definition are now named as resource and resource_detailed. Previous variables for equivalence scale are now named as scale and scale_detailed. The variable for income sharing unit/statistical unit information is now known as reference_unit. The variable for the unit of analysis is now known as sharing_unit. Area coverage variables are now called areacovr and areacovr_detailed. The value All is in some cases, strictly speaking, Representative all or Almost all, but can be used as nationally representative. 7

11 WIID4 does not carry a stand-alone variable for age coverage. It was mostly All for the observations in the previous database version. In cases where it had other values those have now been incorporated into the variable for population coverage. Population coverage information is now provided in the variables popcovr and popcovr_detailed. The latter includes age coverage information in certain cases. For regional information, in addition to the United Nations geoscheme (See United Nations (2016) and Appendix B for more information) variables region_un and region_un_sub, we now also provide the World Bank classification with the variable region_wb. A new variable gdp_ppp_pc_usd2011 is introduced. The values for this variable derive from the World Development Indicators by the World Bank. Gross Domestic Product (GDP) is converted to United States Dollars (USD) using purchasing power parity rates and divided by total population. Data are in constant 2011 United States Dollar (USD). Population variable population is now taken from the United Nations authored World Population Prospects instead of the Penn World Tables. For some historical entities we take the values from the French Institute for Demographic Studies (INED), for Kosovo from the World Development Indicators and for West Bank and Palestine from its statistical authority. In addition to the quality variable, we now provide also a computed variable quality_score, which is a first effort to systematically evaluate the observations. This aims at giving a sense of how much information is provided by each observation, under the understanding that the more information we have about the survey and methodology used to produce the estimates, the better. It also considers how close the estimates are from the standard ones used in the literature. It does not make any consideration, however, about the quality of the survey or the methodology. It is envisioned that together these two quality variables will enable the user to make better informed decisions, if they want to rule out some observations. Variable source has been edited substantially. It is now a general source type indicator with numerical values. The full information for the source is now given in the variable source_detailed. This variable has been cleaned extensively. source_comments remains the same, but in many cases information from it has been moved into the new variable survey, which contains the name of the originating survey for many observations. Database format The data are available in two formats, as an Excel file and as a Stata file. The dataset was prepared using Stata version 15, and the users of earlier version of the software need to do the following: install -use13- by typing in Stata's command prompt: ssc install use13 and then use the use13 command instead of the use command to open the data. 8

12 Documentation The documentation of the database consists of three parts: 1. the documentation of the data in the database itself 2. this user guide 3. country information sheets (these will be subsequently updated). Documentation in the database itself In the database itself, the user is informed about the coverage of the surveys underlying the observations, the income sharing unit, the unit of analysis and the equivalence scale, the income concept and the source and survey used (for details on the variable please refer to the variable list below). The following income/consumption/expenditure concepts are the ones that are mainly used: Net income/disposable income: This label is given if the income concept more or less corresponds to the one specified by the Canberra Group. Even if this label is given, some items might be badly covered. For example, it is not always clear whether in-kind incomes are included or not. Often some in-kind incomes are covered but not home production. Sometimes non-labour incomes are asked in one question that lumps together transfers and income from property. The country-specific documentation and the quality rating give an indication if the income concept is acceptable. Monetary disposable income: This label is given if there is a strong indication that inkind incomes, imputed rents and home production are not included and that the taxes are deducted from the incomes. Gross income: This label is given if the income concept more or less corresponds to the one specified by the Canberra Group before the deduction of taxes and social contributions. The same comments as for the disposable incomes apply. Monetary gross income: This label is given if there is a strong indication that inkind incomes, imputed rents and home production are not included and that the taxes are not deducted from the incomes. Market income, factor income and primary income: This label includes employee income, income from self-employment and property income. Market income also includes private pensions. Earnings only refer to employee income and income from self-employment. A distinction between net and gross earnings has been made. Earnings (without a notion of gross or net) indicates that we do not know whether taxes have been deducted. Income: This label is given if we do not have any information about the income concept from the source (or from some other sources). This means that the income concept might 9

13 include earnings only, monetary incomes only, or it might be net or gross of taxes. Sources not including a definition of the income concept are accepted only if the source is one of the big income distribution compilations or if no other estimates are available for that country and year. Consumption: This label is given if there is a strong indication that the use value, rather than the purchase value of durables is included or if durables are completely excluded. In addition, fines and taxes should not be included in the aggregation. Expenditure This label is given if we know that durables are included with their purchase value and/or taxes and fines are included. This label is also given if we do not have information about the treatment of durables. It is important to note that the distinction between gross and net incomes is sometimes problematic. For example, this is a well-known problem in many Latin American surveys. The issue is that some questionnaires tend to implicitly request gross income, while there is the belief that people paying direct taxes (the formal sector) might actually be reporting take-home wages. For the informal sector, there is basically no difference between net and gross. For this reason, in some cases in which this problem is identified and the source is not clear about whether income is gross or net, income is labelled as net/gross to indicate this ambiguity. In other cases, it is possible that income, even if labelled as net or as gross, still has the same problem. The following income sharing units (variable sharing_unit) are used (mainly): Household: There are variations in the definitions. A broader definition defines the household as covering people who share a dwelling, a more restrictive definition those who share a dwelling and who share resources. Tax unit: The definition depends on the tax laws but is often close to nuclear family. Sometimes children age 18 or over living with their parents are treated as separate tax units. Person: Indicates that the data are collected on the individual level which is in general the case in earnings surveys. The unit of analysis (variable reference_unit) is either household or person. If the unit of analysis is household it means that the size of the households and the needs of different sized households have not been taken into account. If the unit is person it means that the needs of different sized households have been taken into account. The equivalence scale (variable scale) captures the way in which the resource levels of economic units are converted into the resource levels of the population units when equal sharing is assumed. `No adjustment is recorded when the population unit is the same as the economic unit. But various options are possible when as is often the case the original data refer to households but the desired income distribution is defined over individuals. If household needs rise in proportion to household size, then it is appropriate to assign household 10

14 resources per capita to each household member, assuming equal sharing, as is frequently done in the WIID data. At the other extreme, making no adjustment and assigning total household resources to each household member implicitly assumes that additional household members do not increase needs (and there is equal sharing again). Empirical evidence suggests an intermediate position: household needs rise with size, but not in proportion due to economies of scale in consumption. Household equivalence scales reflect this evidence, but differ across time and place, perhaps reflecting differences in household technology and spending patterns but also no doubt due to estimation methods. Thus, there are many different equivalence scales. The four general scales that are used are: Household per capita Household size Square root Household size 0.5 OECD scale Modified OECD scale 1+0.7*n of additional adults + 0.5*n of children 1+0.5*n of additional adults + 0.3*n of children If the variation in equivalence scales used in different cases reflects genuine differences in technology or spending patterns, then that variation is not a concern. But even if the variation does not have such a justification, from the viewpoint of WIID users, the multiplicity of equivalence scales is probably a distraction rather than useful information. So, we group them all together and distinguish only three categories for the scale variable: per capita, equivalized or no adjustment. Note that if a per capita or equivalized scale is applied to household resources, then the population unit must be the individual. Country information sheets In the country information sheets, we have summarized all the relevant documentation that has been available to us about the sources and the surveys used. The sheets start by indicating the sources used and go on to describe the surveys. The years mentioned after the survey names indicate the years of the survey available to us, not the general availability of the survey. To understand the link between the country information sheets and the database it may be useful to check the variable Source Comments in the database. This column will, in most cases, indicate the name of the survey used for a particular estimate. The surveys indicated in this column are described in the sheets. We provide details about the survey coverage, sampling and income/consumption concepts, and if information was available on how the estimates were calculated in the source (column Source1 in the database), we also report that. The country information sheets will often give an impression of how consistent the time series are within sources and countries. The database is increasingly auto-explanatory, so that users in general do not need to read the country information sheets. These country information sheets will, however, be progressively updated. 11

15 Quality rating To give guidance in the use of the database, quality ratings were given to the observations. This was not an easy task because of the heterogeneity of the estimates and the difficulty to decide where to draw the line between high and low quality estimates. The lack of documentation for especially older observations is also a major problem. Criteria used We have used three criteria to evaluate the quality of a data point: 1. Whether the concepts underlying the observations are known or not In principle, this should be evident. In practice, it is far from always the case. Especially in older sources, it is often unclear what the income receiving units and the income concepts are. 2. The coverage of the income/consumption concept The concepts as defined in the most preferred set of underlying definitions have been relied on (see Table 1). For most developed countries, estimates based on monetary incomes have been accepted since the exclusion of in-kind incomes and home production do not have a major effect on the income distribution. The exclusion of imputed rents does have some impact but since estimates are often not available, we have accepted the exclusion. In the case of earnings surveys, income concepts based on earnings are naturally accepted; in the case of household surveys not. This is because earnings do not give a complete picture of the household income. The exception is if the source reports estimates based on several different income concepts to illustrate the difference in inequality among different concepts. Deviations from the preferred income concept are if possible documented in the county information sheets. 3. The survey quality A long list of desirable features could be pointed out, but in practice, coverage issues, questionnaires and data collection methodology were paid attention to. In many cases, the documentation available was insufficient to judge quality for even these issues. We often used additional sources to get information about the surveys. Concerning coverage issues, we do not demand that the coverage should be national. Coverage is not necessarily a quality question, but about what is being measured. A rural household survey cannot be considered of bad quality because it covers rural areas only. The most important thing is that we know the survey coverage, so that rural or urban surveys are not taken for being national ones. Surveys covering very limited areas however are not acceptable, since they do not serve the purpose of the database. Attention was also paid to the exclusion of some special groups, such as households above a certain income threshold only living on charity. Questionnaires or diaries need to have a sufficient level of income or expenditure detail to be acceptable. The data collection methodology is especially important for expenditure surveys and in 12

16 countries where a large proportion of the population works in the informal sector with infrequent incomes. In these cases, too long a recall period leads to considerable measurement errors. For expenditure surveys, diaries must be kept or especially in case of illiteracy frequent visits must be made to the households. Expenditure surveys collected in one single interview or with long recall periods were not considered to be of acceptable quality. Final rating These considerations resulted in the following quality rating: High quality refers to observations where both (a) the underlying income or consumption concepts are known and (b) the quality of the income or consumption concept and the survey are satisfactory according to the criteria outlined above. Average quality refers to observations where either (a) the underlying income or consumption concept or else (b) the quality of the income concept and the survey are unknown or unsatisfactory. The country information sheets will often indicate the specific problems. Low quality indicates observations where both the income or consumption concept and the survey quality are unsatisfactory Not known is the label we attach to observations for which income concept and the survey quality are both indeterminate due to insufficient information. This rating is more common for older observations due to poor documentation. Note that the quality assessment is intended as guidance for users, not as a recommendation that users discard observations not judged to be high quality. While the other observations do not satisfy the rather strict conditions that we have applied, they will still be useful in most applications. Quality score In addition to the quality variable, we now provide also a computed quality score. This aims at giving a sense of how much information is provided by each observation, under the understanding that the more information we have about the survey and methodology used to produce the estimates, the better. It also considers how close the estimates are from the standard ones used in the literature. It does not make any consideration, however, about the quality of the survey or the methodology. We award points to the observations based on their attributes in the following way (maximum is 13 points) Gini coefficient is available (1) Resource concept: o Consumption, Income (net), Income (gross), Monetary income (gross), Monetary income (net) (5) o Income, Monetary income, Market income (3) o Factor income, Primary income, Taxable income, Earnings (1) Equivalence scale: o Per capita or equivalized (3) 13

17 o No adjustment (2) Area coverage: o All, Urban, Rural (1) Population coverage: o All (1) Distributional share information: o All of d1-d1 are available (2) o All of q1-q5 are available (at least one of d1-d10 is missing) (1) Some final guidelines The user is advised to: 1. pay attention to definitional differences as documented in the database 2. consult the country sheets concerning information about individual countries (these will be made available at the WIID website) 3. keep in mind that sources which adapt different income concepts or different statistical units cannot be combined or compared unless data corrections and adjustments are introduced 4. keep in mind that data points with similar definitions are not automatically comparable since differences in survey methodology might impair the comparability 5. report in their research paper which series of Ginis they used from the WIID; i.e. provide knowledge of their algorithms of data selection to make sure readers understand which observations were used 14

18 Referring to the WIID Please refer to the present WIID (along with the version date) as: UNU-WIDER, World Income Inequality Database (WIID4) List of variables Variables used in WIID4 id Identifier country c3 c2 year gini_reported q1-q5 d1-d10 bottom5 and top5 resource resource_detailed scale scale_detailed sharing_unit reference_unit areacovr areacovr_detailed popcovr popcovr_detailed region_un region_un_sub region_wb eu oecd Country/area 3-digit country code in ISO alpha-3 format 2-digit country code in ISO alpha-2 format Year. Note that when a survey continues for more than a year, the year when it is finished is considered Gini coefficient as reported by the source (in most cases based on microdata, in some older observations estimates derive from grouped data) Quintile group shares of resource Decile group shares of resource Bottom five and top five percent group shares of resource Resource concept Detailed resource concept Equivalence scale Detailed equivalence scale Income sharing unit/statistical unit Unit of analysis, indicates whether the data has been weighted with a person or a household weight Area coverage. The land area which was included in the original sample surveys etc. Detailed area coverage Population coverage. The population covered in the sample surveys in the land area (all, rural, urban etc.) which was included Detailed population coverage, including age coverage information in certain cases Regional grouping based on United Nations geoscheme Sub-regional grouping based on United Nations geoscheme Regional grouping based on World Bank classification Current EU member state Current OECD member state 15

19 incomegroup mean median currency reference_period exchangerate mean_usd median_usd gdp_ppp_pc_usd2011 population revision quality quality_score source source_detailed source_comments survey World Bank classification by country income Survey mean given with the same underlying definitions as the Gini coefficient and the share data Survey median given with the same underlying definitions as the Gini coefficient and the share data Currency for the mean and median values. If the reference is US$2011PPP it means that the currency is in 2011 US dollar per month, with purchasing power parity applied on it. Time period for measuring mean and median values Conversion rate from local currency units (LCU) to United States Dollars (USD) Mean measure in United States Dollar (USD) Median measure in United States Dollar (USD) Gross Domestic Product (GDP) is converted to United States Dollars (USD) using purchasing power parity rates and divided by total population. Data are in constant 2011 United States Dollar (USD) Population of countries from the UN population prospects Indicates the time of the revision when the observation was included to the database Quality assessment Computed quality score Source type Source from which the observation was obtained Additional source comments Originating survey information 16

20 Cumulative % of income or expenditure Glossary Lorenz curve and the Gini coefficient 100 A 0 Cumulative % of reference units 100 A straightforward graphical interpretation of the Gini coefficient is in terms of the Lorenz curve, which is the thick curve in the figure above. The horizontal axis measures the cumulative percentage of the population, whose inequality is under consideration, starting from the poorest and ending with the richest. The vertical axis measures the cumulative percentage of income (or expenditure) associated with the units on the horizontal axis. In case of a completely egalitarian income distribution in which the whole population has the same income, the Lorenz curve would be the dashed 45-degree line. When incomes vary within the population, the poor population has a proportionately lower share of income compared with the rich population, and the Lorenz curve may look like the above thick curve below the 45-degree line. As inequality rises, the thick curve moves towards the bottom right-hand corner. The Gini coefficient is the area A between the 45-degree line and the Lorenz curve, divided by 1/2, the total area under the 45-degree line. The Gini coefficient may be given as a proportion or percentage. From this it is clear that the Gini coefficient will be equal to 0 when the distribution is equal. If the society's total income accrues to only one person/household unit, leaving the rest with no income at all, then the Gini coefficient approaches 1, or 100%. Equivalence scales One complication posed by use of the household as the statistical unit is that households vary in size and composition and such differences between households mean that their relative needs will be different. For example, a large household will have a lower standard of living from the same income as that received by a small household, all other things being equal. Costs of household members also differ according to their age, student status, labour force status and so on. 17

21 Equivalence scales are designed to adjust income/consumption to account for differences in need due to differences in household size and composition. The most basic of such adjustments is to calculate household income/consumption per member to adjust total incomes/consumption according to the number of people in the household. But such an adjustment ignores economies of scale in household consumption relating to size and other differences in needs among household members, in particular differing needs according to the age of both adults and children. There is a wide range of equivalence scales in use in different countries and by different organisations. All take account of household or family size: in many scales this is the only factor, whilst in those taking into account other considerations it is the factor with greatest weight. Equivalence scales are usually presented as income/consumption amounts, or ratios of amounts, needed by households of different size and structure. Thus, if a one-person household needs one unit of income/consumption to maintain a given level of living, a twoperson household may need 1.7 units, and a three-person household 2.2 units. There are two basic approaches to construction of scales: those which use the expert knowledge of social scientists and others, and those which are developed empirically based on analysis of survey data. (Citation from the Canberra Group Report, 2001, p.40) Quintile, decile, percentile group shares The quintile group shares express the share of total income going to each fifth of the population ordered according to the size of their incomes. In WIID4, these shares are expressed as percentages of total income. The first quintile group includes the poorest 20% of the population, while the fifth quintile includes the richest 20%. Deciles divide the population into ten groups and percentiles into one hundred groups. Unit record data/microdata Data that contain information on unit level from the survey; in the case of income or consumption distribution data the units is most often the household or the members of the household. If, for example, 8,000 households took part in a survey, the unit record data include all 8,000 households or household members. Grouped data This is data available in some kind of grouped form, for example the number of persons in income classes or quintile/decile group data. Imputed rents for owner-occupied dwellings This is the imputed value of the services provided by a household s residence, after deduction of expenses, depreciation and property taxes. Home ownership may offset other costs and is therefore important. The main problem is the accurate measurement of imputed rent. The value of the rent of owner-occupied dwellings should in principle be the market rental value of an exactly similar house (Canberra Group Report, 2001, pp.63,120). Home consumption Value of goods produced and consumed within the households, less expenses incurred in production. Inclusion of this item is particularly important in countries where subsistence agriculture is significant (Canberra Group Report, 2001, p.120). 18

22 Appendix A: Codebook The value labels for numeric variables are listed in below. In addition to the grouped variables (e.g. resource), we provide the detailed variables (e.g. resource_detailed) that contain the full information. The values of the detailed variables match to the grouped variables with the first digit: For example, 202 Monetary income in the resource_detailed is 2 Income (net/gross) in resource. resource Value Label 1 Income (net) 2 Income (net/gross) 3 Income (gross) 4 Consumption 5 Earnings resource_detailed Value Label 101 Income, net 102 Monetary income, net 103 Monetary income, net (excluding property income) 201 Income, net/gross 202 Monetary income 301 Income, gross 302 Monetary income, gross 401 Consumption 501 Earnings 502 Earnings, gross 503 Earnings, net 504 Factor income 505 Market income 506 Primary income 507 Taxable income, excluding property income 508 Taxable income, gross 509 Taxable income, gross (including deductions) 510 Taxable income, net 601 Income/consumption 19

23 scale Value Label 1 Per capita 2 Equivalized 3 No adjustment scale_detailed Value Label 101 Per capita 102 Family per capita 103 Head of household 104 Tax unit per capita 201 Equivalized McClements scale revised Jensen scale 204 Census family equivalized, square root 205 Economic family equivalized, square root 206 Family equivalized 207 Family equivalized, OECD 208 Family equivalized, national scale 209 Family equivalized, social assistance 210 Family equivalized, square root 211 Household equivalized, social assistance 212 National scale 213 OECD 214 OECD-modified 215 Tax unit equivalized, square root 301 No adjustment sharing_unit Value Label 1 Household 2 Family 3 Tax unit 4 Person reference_unit Value Label 1 Person 2 Household 3 Family 20

24 4 Tax unit areacovr Value Label 1 All 2 Rural 3 Urban 4 Part areacovr_detailed Value Label 101 All 102 All, excl. Abkhasia and Tskhinvali 103 All, excl. Costa Rural, Selva Rural and Selva Urbana (30% of the population) 104 All, excl. East Timor 105 All, excl. East-Central State 106 All, excl. Transnistria 107 All, excl. West Irian and East Timor 108 All, excl. West Irian, East Timor and Maluku 109 All, excl. eight districts in the north and the east (15% of the population) 110 All, excl. nomadic areas 111 All, excl. northern and eastern provinces 112 All, excl. seven districts 113 All, excl. some special areas (4% of the population) 114 Continental Portugal 115 Main island 116 With rural north 117 Without rural north 201 Rural 202 Agricultural sector 203 Four rural areas 204 Rural, excl. seven districts on national level 301 Urban 302 All, mainly urban areas 303 Capital 304 Cities 305 Cities (n=16) 306 Cities (n=17) 307 Cities (n=4) 308 Cities (n=7) 309 Cities (n=8) 310 Metropolitan area 21

25 311 Nonagricultural sector 312 Paramaribo and Wanica 313 Urban, excl. Western Province 314 Urban, excl. metropolitan area 315 Urban, excl. seven districts on national level 401 Estate sector 402 Four areas 403 Java 404 Nonmetropolitan area 405 Peninsular Malaysia 406 Six northern provinces 407 Three cantons 408 East Germany 409 West Germany popcovr Value Label 1 All 2 Economically active 3 Specific categories popcovr_detailed Value Label 101 All 201 Economically active 202 Employed 203 Family units with earnings 204 Households with earnings 205 Income recipients 206 Taxpayers 301 Agricultural households 302 All excl. some private sector employees 303 All, aged All, aged All, excl. farmers 306 All, excl. fishery hhs and farm hhs with very small land holdings 307 All, excl. foreign-head hhs + hhs with net income DM > All, excl. foreign-headed hhs 309 All, excl. foreign-headed hhs + hhs with net income DM >= All, excl. foreign-headed hhs + hhs with net income DM >= All, excl. foreign-headed hhs + hhs with net income over a certain limit 312 All, excl. hhs with net income DM >

26 313 All, excl. hhs with wives aged All, excl. households depending entirely on charity 315 All, excl. nomadic people (30% of the population 316 All, excl. pensioner-headed households 317 All, excl. pensioners 318 All, excl. self-employeds in the high income brackets 319 All, excl. single-member households 320 All, excl. very high income households 321 All, unclear if inclusive of nomadic people (30% of the population) 322 Employed, > 10 employees 323 Employed, > 25 employees 324 Employed, > 5 employees 325 Employed, African males 326 Employed, aged Employed, aged Employed, excl. entrepreneurs and farmers, >= 3 employees 329 Employed, excl. independent farmers, persons employed in crafts and trade 330 Employed, excl. private enterprises, >= 20 employees 331 Employed, excl. self-employeds 332 Employed, excl. small enterprises 333 Employed, excl. small enterprises and cooperatives 334 Employed, excl. small private enterprises 335 Employed, full-time 336 Employed, full-time employees in the public sector 337 Employed, full-time, >= 100 employees 338 Employed, full-time, >= 20 employees 339 Employed, full-time, >= 25 employees 340 Employed, full-time, >= 25 employees, some sectors >= 100 employees 341 Employed, full-time, >= 50 employees 342 Employed, full-time, excl. self-employeds and farmers 343 Employed, multi-member households 344 Employed, private sector 345 Employed, public sector 346 Employed, public sector, excl. social organizations 347 Employed, socialized sector 348 Employed, socialized sector, > 5 employees 349 Employed, state and cooperative sector 350 Employed, state sector 351 Employee households 352 Estate sector 353 Excl. self-employed households 354 Households where head employed or inactive 355 Households with positive or zero taxable incomes 23

27 356 Income recipients, aged Income recipients, public sector 358 Males, aged Mostly families of state sector and collective farm employees 360 Non-agricultural households 361 Non-agricultural multi-member households 362 Non-estate sector 363 Non-farm population 364 Omani 365 Qatari 366 Saudi 367 Self-employed households 368 Taxpayers, Jewish 369 Taxpayers, aged Taxpayers, aged Taxpayers, aged Taxpayers, permanently employed and self-employed 373 Wage earners 374 Workers 375 Workers, state and cooperative sector region_un Value Label 1 Americas 2 Europe 3 Africa 4 Asia 5 Oceania region_un_sub Value Label 101 Northern America 102 Central America 103 Caribbean 104 South America 201 Northern Europe 202 Western Europe 203 Eastern Europe 204 Southern Europe 301 Northern Africa 302 Western Africa 303 Middle Africa 24

28 304 Eastern Africa 305 Southern Africa 401 Western Asia 402 Central Asia 403 Southern Asia 404 Eastern Asia 405 South-eastern Asia 501 Australia and New Zealand 502 Micronesia 503 Melanesia 504 Polynesia region_wb Value Label 1 North America 2 Latin America and the Caribbean 3 Europe and Central Asia 4 Middle East and North Africa 5 Sub-Saharan Africa 6 South Asia 7 East Asia and the Pacific eu Value Label 0 Non-EU 1 EU oecd Value Label 0 Non-OECD 1 OECD incomegroup Value Label 1 High income 2 Upper middle income 3 Lower middle income 4 Low income reference_period Value Label 25

29 1 Year 2 Month 3 Week 4 Day quality Value Label 1 High 2 Average 3 Low 4 Not known source Value Label 1 Luxembourg Income Study 2 Eurostat 3 SEDLAC 4 United Nations 5 National statistical authority 6 OECD 7 World Bank 8 Research study 9 Other international organizations 26

30 Appendix B: United Nations geographical sub-regions Source: Wikimedia Commons. GNU Free Documentation License, Version 1.2; created by B. Arnold; 27

31 Appendix C: World Bank regional classification Source: Wikimedia Commons. GNU Free Documentation License, Version 1.2; created by Hanteng; 28